# Compute Technicals

> Compute standard technical indicators (RSI, SMA, EMA, MACD, Bollinger) on an OHLCV frame; tested against canonical references.

- Skill: `jcaiagent7143-ui/compute-technicals` (Agent Skill)
- Install (CLI): `npx skillmds@latest add jcaiagent7143-ui/compute-technicals`
- Raw SKILL.md: https://api.skillmd.com/api/skills/jcaiagent7143-ui/compute-technicals/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: jcaiagent7143-ui (https://skillmd.com/u/jcaiagent7143-ui)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/jcaiagent7143-ui/compute-technicals

---


# Compute technicals

## Canonical formulas (no hand-waving)

| Indicator | Definition |
|---|---|
| **SMA(n)** | Simple moving average of close over n bars |
| **EMA(n)** | Exponential moving average with α = 2/(n+1) |
| **RSI(14) — Wilder** | `RS = avg_gain_14 / avg_loss_14` (Wilder smoothing, not simple average); `RSI = 100 - 100/(1+RS)` |
| **MACD** | EMA(12) − EMA(26); signal = EMA(9) of MACD |
| **Bollinger(20, 2)** | SMA(20) ± 2 × stddev(20) |

The **Wilder smoothing** distinction matters — many tutorials use a simple average and get RSI values off by a few points. The first 14 bars use the simple average to seed; from bar 15 onward, each step is `(prev * 13 + new) / 14`.

## Steps

1. **Validate inputs** — frame must have `Close`. For volume-based indicators, `Volume` too.
2. **Compute in-place on a copy** — never mutate the caller's frame.
3. **Append columns** named exactly per canonical lib conventions (e.g. `RSI_14`, `SMA_50`, `SMA_200`, `MACD`, `MACD_signal`).
4. **NaN for warmup periods** — don't forward-fill the first 13 RSI values.
5. **Test against a known fixture** — at least one indicator should match a published value (e.g. RSI of a flat series = 50; RSI of monotonically increasing close → 100).

## Failure modes to avoid

- Using simple average instead of Wilder smoothing for RSI.
- Forward-filling warmup NaN — gives the screen false signals at the start of the window.
- Computing RSI on intraday bars but labeling it as daily.
- Adding pandas-ta / ta-lib as a dep for indicators you can write in 15 lines and test.